[cnicholas]

Closes #143

Summary

• What: Xbar/S chart_table shows subgroup and n whatever the by=; Xbar/S chart data carries n.
• Why: with a by= other than time alone (pooled subsets, all factors, factors without time, one factor) the table dropped both columns. A row couldn't be traced to its subgroup and stepped limits went unexplained. The app shows these tables directly.
• Scope: the Xbar/S aggregation already counts each subgroup; it now keeps that n in the chart data instead of dropping it, and names its label column in metadata (group_col). The table reads both, and the per-cell re-join that could never size a pooled subgroup (_join_n_on_composite_group) is gone.

Contract / Invariants (must remain true)

• Default behavior unchanged: limits, centres and statistics byte-identical on 104 Xbar/S requests over T100 PM SDS 1 and 3 (snapshot vs main)
• Residuals unaffected: no residual or design-state change
• Row/index alignment preserved: row count and order unchanged
• Lane boundaries preserved: untouched
• Output schema compatible: additive — Xbar/S chart data gains n; metadata gains group_col (CHANGELOG "Changed")

Behavior Changes (explicit)

• New semantics:
  • get_chart('Xbar'/'S') has n after the subgroup label (after stratify_col, group_col on stratified charts); Excel chart sheets too
  • Every Xbar/S chart_table has subgroup, n; a pooled subgroup's n covers all its cells
  • n is the count the limits rest on. On residual charts whose limits come from R2 (R1, R3-R5) in a design with singletons, the stream's first cell now shows one fewer than its cell size, as its S and limits use (main showed the cell size there, which disagreed with its own limits; one row on PM SDS 3)

Key Design Decisions

• n source: the aggregation's own count (Chris's call), not a re-join from per-cell sizes
• Label column: declared by the chart (metadata['group_col']) rather than guessed from column names; stratified and X/mR tables keep the rsg fallback
• NaN policy: n counts the limits column's non-missing values, as the aggregation always has
• n_mode='average': n stays the actual sizes; the average remains in N / metadata['n_avg']

Risks & Mitigations

• P0 risks: a limits change hiding behind the schema change
• Mitigations: snapshot comparison against main (limits/stats identical, only n added); golden masters regenerated only after checking the diff was exactly the n column and group_col key

Tests

Added / Updated

• TestXbarSSubgroupAndN (test_result_tabular.py) — 23 cases on T100: every by= variant has subgroup and an independently counted n (Xbar n >= 1, S n >= 2); solo S == companion S; default and stratified tables equal cell sizes; R5 n follows R2's count; n_mode='average' keeps actual sizes; n sits after the label in chart data
• Golden masters for paired_xbar_s, residual_chart, stratified_balanced, stratified_uneven regenerated (n column + group_col only)

Regression Focus (must fail if broken)

• 16 of the new cases fail on main's code; the 7 regression guards pass on both

Manual Verification

• pytest tests/ — 2567 passed, 7 skipped
• python validation/e2e_bishop_report.py — 1547 passed, 0 failed, 4 pending (14 runs); docs/reference/validation.md unchanged
• App fast tests against the local library (PYTHONPATH) — 193 passed
• Survey of by= variants on make_design(1): every Xbar/S table shows subgroup and n

Notes

• Docs: chart_table docstring, api.md, CHANGELOG Fixed + Changed; stored outputs of the cells printing Xbar/S chart data refreshed (xbar-s-analysis cells 11, 12, 24, 25; quickstart cell 13)
• Follow-ups (not in this PR): none